The most common way I see a company fail at AI search is not bad content or the wrong strategy. It is quitting in week three.
Here is how it goes. A founder reads that AI engines are the new top of the funnel, takes it seriously, and ships a handful of genuinely good pages built to be cited. Two weeks later they run the queries, see their name nowhere, and quietly conclude the whole thing does not work for a company their size. They were right about the opportunity and wrong about the clock. The work was fine. They just checked the scoreboard during the part of the game where the score does not move yet.
So the real question is not whether AI search works. It is how long it takes, and what the waiting actually looks like. The honest answer has two parts, because there are two different clocks running.
Clock one: how long until you can be cited at all
Before any engine can recommend you, it has to find your page, trust it enough to use, and have a path to surface it. How long that takes depends almost entirely on how the specific engine works, and the engines do not agree.
Some engines retrieve live. When you ask Perplexity, Google’s AI answers, or Copilot a question, they go and fetch current sources at the moment of the query. For these, the gap between publishing a strong page and being eligible for citation is short, often days to a few weeks, roughly as fast as the page gets crawled and indexed. You can earn a citation on a retrieval engine while the page is still new.
Some engines answer from memory. The default ChatGPT experience often answers from what the model already absorbed during training, not from a live fetch. That knowledge updates on training refreshes, which happen on the order of months, not days. You can publish the best page in your category and the memory-based answer will not know about it until the next refresh, no matter how good it is. I went deep on this split in how AI engines decide what to cite, and it is the single biggest reason “how long does it take” has no single number.
So the first realistic milestone is this: on the retrieval-based engines, expect your first citations within a few weeks of publishing genuinely citable content. On the memory-based ones, expect a longer wait tied to refresh cycles you do not control. Anyone who promises you fast citations everywhere is either talking about one engine or selling you something.
Clock two: how long until it actually adds up
Being eligible to be cited is not the same as being cited at volume. This is the clock that catches people out, because it does not move the way they expect.
AI citations do not ramp in a straight line. They compound. For the first stretch the numbers are small and look flat, because only a few of your pages have been picked up and the engines are still deciding whether to trust you on each question. Then, as more pages get pulled and the engine starts preferring you as a source, the same work that produced almost nothing starts producing a lot, and the daily count bends upward on its own.
I watched this happen on a real engagement recently. For roughly the first six weeks the citation count was close to flat, the kind of flat that makes you want to change everything. Then over the next three weeks the daily number roughly tripled, off the same content plan, with no sudden new tactic. Nothing changed except that the slow part finished. If that company had judged the program at week four, they would have killed it right before it worked.
That is the shape to expect: a flat stretch that feels like failure, then compounding. The flat part is not the program failing. It is the program loading.
So what is the honest timeline
Putting both clocks together, here is what a realistic engagement actually looks like, with the caveat that your domain’s existing trust and how contested your category is will move these.
First citations on retrieval-based engines: a few weeks. Meaningful, repeatable citation volume across several pages: two to three months. The compounding bend, where the curve starts working for you: somewhere after that, and then it keeps going as long as you keep feeding it. The memory-based engines come in on their own slower schedule on top of all this.
None of that is fast in the way a paid ad is fast. It is also not the vague “give it six months” that people say to avoid answering. It is a specific shape with a slow front and a compounding back, and knowing that shape is what lets you hold your nerve through the part that looks like nothing.
What I got wrong
I used to talk about timelines in averages, “expect movement in a couple of months,” and it set people up to misread the flat stretch as failure. On one engagement I almost let a client narrow their content plan at week four because the citation line was barely off zero. The plan was working, the crawl-and-trust phase just had not paid out yet. We held it, and the curve bent two weeks later.
Now I do the opposite. I tell people the first weeks will look flat, name it before it happens, and point them at leading indicators instead of the headline number. The fix was not a better tactic. It was setting the expectation honestly so the slow part did not read as a verdict.
How to survive the flat part
The trap is judging the program by citation volume during the exact window when citation volume is the wrong thing to watch. So watch the earlier signals instead.
Are your new pages getting crawled. Are they starting to appear on the fast, retrieval-based engines even before the slower ones notice. Is the engine quoting you on a few questions even if the totals are small. Those are the signs the machine is doing what it is supposed to, and they show up weeks before the volume does. The way to track them on purpose, rather than by vibes, is the citation protocol in how to measure AI search visibility, and the reason retrieval mode should shape what you publish first is in how to build an AI search content plan.
If those leading signals are moving, the flat headline number is not a problem. It is the expected first chapter.
Why this matters
The companies that win AI search are usually not the ones with the cleverest tactics. They are the ones who did reasonable work and then did not flinch during the slow part. The barrier to entry here is mostly patience, which is good news, because patience is a thing you can choose and your competitors mostly will not.
If you want to know where you actually stand right now, and whether the quiet you are seeing is the loading phase or a real problem, that is the first thing a paid audit tells you, and the full approach is on the methodology page.